NVIDIA’s Jensen Huang: Demand ‘Parabolic,’ China Trip

NVIDIA CEO Jensen Huang made three significant moves in a single week — declaring AI infrastructure demand “utterly parabolic” at Dell Technologies World, joining President Trump’s China summit at the president’s personal request, and backing British startup Ineffable Intelligence to pursue reinforcement-learning-based AI systems. The cluster of events, spanning May 12–14, 2026, underscores how central Huang and NVIDIA have become to both the technology industry and U.S. trade policy.

Huang at Dell Technologies World: Demand Is ‘Utterly Parabolic’

Speaking Monday at Dell Technologies World, Huang told the audience that AI has crossed into practical utility — and that adoption is accelerating faster than at any prior point in the industry’s history. “We’ve now arrived at the era of useful AI, which is the reason why demand is going parabolic, utterly parabolic,” Huang said, according to the NVIDIA AI Blog. “What took months now takes weeks. What took weeks now takes days.”

Dell chairman and CEO Michael Dell set the context before Huang took the stage, projecting that worldwide AI infrastructure spending could reach $3–4 trillion by 2030, with token consumption forecast to grow 3,400% over the same period. Dell noted that 5,000 enterprises — including Eli Lilly, Samsung, and Honeywell — are already running AI workloads on Dell AI Factories built with NVIDIA hardware.

Huang’s appearance was framed around the full NVIDIA product stack, from a deskside Dell Pro Max with GB10 workstation to a Dell PowerRack running the NVIDIA Vera Rubin NVL72. The range signals NVIDIA’s push to capture AI compute spending at every tier of enterprise deployment, not just hyperscale data centers.

Vera Rubin NVL72 and Vera CPU: The Numbers Behind the Claims

The keynote put specific performance figures behind NVIDIA’s newest hardware. According to the NVIDIA AI Blog, the Vera Rubin NVL72 delivers agentic AI inference at one-tenth the cost per token compared to prior-generation systems — a claim aimed directly at enterprises weighing the economics of running large language models at production scale.

The NVIDIA Vera CPU posted two headline numbers:

• 50% faster agent sandbox execution than traditional CPUs
• 3x faster enterprise data query performance

These figures position the Vera CPU not merely as a companion chip to NVIDIA’s GPUs, but as a standalone performance argument for organizations running mixed AI and data workloads. The Vera Rubin NVL72 — a 72-GPU rack-scale system — is designed for inference-heavy deployments where per-token cost, rather than raw training throughput, is the primary economic constraint.

The Vera architecture represents NVIDIA’s first in-house CPU design to reach enterprise deployment at scale, combining Arm-based compute with NVIDIA’s NVLink interconnect fabric to reduce data movement bottlenecks between CPU and GPU.

NVIDIA Backs British Startup Ineffable Intelligence

On Wednesday, May 13, NVIDIA announced a partnership with London-based startup Ineffable Intelligence, focused on building AI systems that learn from experience rather than from static human-generated training data. CNBC reported that Ineffable’s core approach centers on reinforcement learning — training models through iterative feedback loops rather than the supervised learning pipelines that underpin most current frontier models.

“The next frontier of AI is superlearners — systems that learn continuously from experience,” Huang said in the announcement.

The partnership is notable for what it signals about NVIDIA’s strategic positioning. Rather than remaining purely a hardware supplier, NVIDIA is increasingly placing bets on specific AI research directions — effectively shaping which architectures and training paradigms will drive future GPU demand. Reinforcement learning workloads are computationally intensive and poorly served by inference-optimized infrastructure, which could expand NVIDIA’s addressable market if the approach gains traction.

Ineffable Intelligence had not previously disclosed significant funding or partnerships before this announcement. The terms of the NVIDIA partnership were not disclosed.

Jensen Huang Joins Trump’s China Summit

In a development that intersects technology and geopolitics, Huang confirmed Thursday, May 14, that he joined President Donald Trump’s China trip — and that Trump personally asked him to attend. “President Trump asked me to come,” Huang told reporters, according to CNBC.

Early reports had suggested Huang would not be part of the delegation. His inclusion was announced Tuesday, just days before the summit. Huang described the trip as “one of the most important summits in human history.”

The China dimension carries direct commercial weight for NVIDIA. Export controls introduced during the Biden administration restricted sales of NVIDIA’s most advanced chips — including the H100 and subsequent architectures — to Chinese customers, forcing NVIDIA to develop downgraded variants like the H20 for that market. Any shift in U.S.-China trade posture could materially affect NVIDIA’s revenue in a market it previously dominated.

CNBC also noted that analysts are watching closely to see whether Huang addresses China chip policy on NVIDIA’s upcoming earnings call, given the timing of the Xi summit.

What This Means

The week’s events collectively illustrate NVIDIA’s position as something beyond a chip company. Huang is now a participant in U.S. foreign policy summits, a backer of early-stage AI research startups, and the central figure at major enterprise technology conferences — all within 72 hours.

The Vera Rubin NVL72’s one-tenth cost-per-token claim, if it holds in production deployments, would significantly shift the economics of agentic AI at enterprise scale. Cost per token has been the primary barrier preventing large-scale autonomous agent deployment; reducing it by 90% would likely accelerate adoption faster than any performance improvement alone.

The Ineffable Intelligence bet is smaller in immediate commercial terms but strategically coherent: if reinforcement learning produces the next generation of capable AI systems, NVIDIA wants its hardware — and its brand — embedded in that research from the start.

On China, the risks remain asymmetric. Loosened export controls could restore a major revenue channel; tightened controls or a deteriorating diplomatic relationship could further constrain it. Huang’s presence at the summit suggests NVIDIA has decided that direct engagement with that policy process is preferable to watching from the sidelines.

FAQ

What is the NVIDIA Vera Rubin NVL72?

The Vera Rubin NVL72 is a rack-scale AI system from NVIDIA featuring 72 GPUs, designed primarily for inference workloads. NVIDIA claims it delivers agentic AI inference at one-tenth the cost per token compared to prior-generation systems, targeting enterprises running large language models at production scale.

Why did Jensen Huang join Trump’s China trip?

Huang said President Trump personally asked him to attend the U.S.-China summit, which took place in mid-May 2026. NVIDIA has significant commercial interests in China that have been constrained by U.S. export controls on advanced chips, making the diplomatic outcome of the summit directly relevant to the company’s business.

What is Ineffable Intelligence and why is NVIDIA partnering with it?

Ineffable Intelligence is a British AI startup focused on reinforcement learning — training AI models through experience rather than static human-generated data. NVIDIA announced a partnership with the company on May 13, 2026, with Huang framing reinforcement-learning-based “superlearners” as the next major frontier in AI development.